Probability and Random

Processes

STA 3533

Module 6

Networks of Queues
Continue
 Topics
Case Study Solution
Introduction to Queuing

Open queuing networks

Closed queuing networks

 Introduction to Queuing
 Problem 1
A data communication line delivers a block of information
every 10 microseconds. A decoder checks each block for
errors and corrects the errors if necessary. It takes 1 s to
determine whether a block has any errors. If the block has
error, it takes 5 s to correct it, and if it has more than one
error it takes 20 s to correct the error. Blocks wait in a
queue when the decoder falls behind. Suppose that the
decoder is initially empty and that the numbers of errors in the
first 10 blocks are 0, 1, 3, 1, 0, 4, 0, 1, 0, 0.
a. Plot the number of blocks in the decoder as a function of time.
b. Find the mean number of blocks in the decoder.
c. What percentage of the time is the decoder empty?
 Introduction to Queuing
 Solution 1:
1. Interarrivals are constant with
interarrival times = 10 msec
2. Service time
if 0 error 1 sec
1 error 1+5 sec
>1 error1+20 sec

Arrival time 10 20 30 40 50 60 70 80 90 100

Errors 0 1 3 1 0 4 0 1 0 0

Service time 1 6 21 6 1 21 1 6 1 1

Dep. Time 11 26 51 57 58 81 82 88 91 101

 Introduction to Queuing
 a. The plot
N(t)

10 20 30 40 50 60 70 80 90 100 110
TObservation = 100 msec

1 110
 b. 100 10
N (t )dt

1
 [1  6  10  20  3  12  1  10  20  3  2  6  1  1]
100
 0.96
 Introduction to Queuing
 c.
Server is working during 65 msec =  service times
proportion idle time = 1 – 65/100
= 0.35
 Introduction to Queuing
 Problem 2:
 Three queues are arranged in the loop as shown below. Assume that
the mean service time in the queue i is mi = 1/ .
a. Suppose the queue has a single customer circulating in the loop. Find
the mean time E[T] it takes the customer to cycle around the loop.
Deduce from E[T] the mean arrival rate  at each of the queues. Verify
that Little’s formula holds for these two quantities.
b. If there are N customers circulating in the loop, how are the mean
arrival rate and mean cycle time related?

  
 Introduction to Queuing
 Solution:
 a. One customer  no waiting
1
  [T ]  m1  m2  m3   
m1  m2  m3

Little’s Formula 
mi
 [ N i ]  mi   %time customer in queue i
m1  m2  m3
3 3
mi
  [ Ni ]  
i 1 i 1 m1  m2  m3
 1  one customer in system
 Introduction to Queuing
b. Let [T] = mean cycle time per customer, then

Total # in system = N – [T] by Little’s formula

 Open Queuing Networks
Consider a network of K queues in which customers
arrive from outside the network to queue k according to
independent Poisson processes of rate k. We assume
that the service time of a customer in queue k is
exponentially distributed with rate k and independent
of all other service times and arrival processes. We also
suppose that queue k has ck servers.
After completion of service in queue k, a customer
proceeds to queue i with probability Pki and exists the
network with probability K
1   Pki
i 1
 Open Queuing Networks
The total arrival rate k into queue k is the sum of the
external arrival rate and the internal arrival rates:
K
k   k    j Pjk k  1,..., K .
j 1

It can be shown that above equation has a unique

solution if no customer remains in the network
indefinitely.
Such networks are known as Open Queuing Networks
 Open Queuing Networks
The vector of the number of customers in all queues,
N (t )  ( N1 (t ), N 2 (t ), N 3 (t ),..., N K (t )), is a Markow process.
Now Jackson ‘s theorem gives steady state pmf for N(t).
Jackson’s Theorem: if k < ckk, then for any possible
state n= (n1, n2,…,nK ),
P[ N (t )  n]  P[ N1  n1 ]P[ N 2  n2 ]...P[ N K  nK ]

Where P[ Nk = nk ] is the steady state pmf of an M/M/ck

system with arrival rate k and service rate k
 Open Queuing Networks
Jackson’s Theorem states that the numbers of
customers in the queues at time t are independent
random variables.
In addition, it states that the steady state probabilities
of the individual queues are those of an M/M/ck system.
This is an amazing result because in general the input
process to a queue is not Poisson, as was demonstrated
in the sample queue with feedback discussed in the
beginning of this section.
 Open Queuing Networks
 Example: Messages arrive at a concentrator according to a
Poisson process of rate . The time required to transmit a
message and receive an acknowledgment is exponentially
distributed with mean 1/m. Suppose that a message need to
be retransmitted with probability p. Find the steady state
pmf for the number of messages in the concentrator.
 Solution: The overall system can be represented by the
simple queue with feedback shown below.



 


 Open Queuing Networks
The net arrival rate into the queue is  =  + p, that
is,

 
1 p

Thus, the pmf for the number of messages in the

concentrator is

P[ N  n]  (1   )  n n  0,1,...
where    /    /(1   ) 
 Open Queuing Networks
 Example: New programs arrive at a CPU according to a Poisson
process of rate  as shown in below fig. A program spends as
exponentially distributed execution time of mean 1/1 in the CPU.
At the end of this service time, the program execution is complete
with probability p or it requires retrieving additional information
from secondary storage with probability 1-p. Suppose that the
retrieval of information from secondary storage requires an
exponentially distributed amount of time with mean 1/2. Find the
mean time that each program spends in the system.

CPU p 


 p 
 Open Queuing Networks
 Solution: the net arrival rates into the two queues are
1    2 and 2  (1  p )1
 Thus  (1  p)
1  and 2 
p p

 Each queue behaves like an M/M/1 system so,

1 2
E[ N1 ]  and E[ N 2 ] 
1  1 1  2

 Where and . Little’s formula then gives the

mean for total time spent in the system:
E[ N1  N 2 ] 1  1 2 
E[T ]     .
  1  1 1   2 
 Close Queuing Networks
 In some problems, a fixed number of customers, say I,
circulate endlessly in a closed queuing network.
 For Example, some computer system models assume that at
any time a fixed number of programs use the CPU and input/
output (I/O) resources of a computer as show in below figure.

CPU 
p
 
 Close Queuing Networks
 We now consider queuing networks that are identical to the
previously discussed Open Network except that the external
arrivals rates are zero and the networks always contain a fixed
number of customers I. We show that the steady state pmf for
such system is product from but that the states of the queues are
no longer independent.
 The net arrival rate into the queue k is now given by
K
k    j Pjk k  1,..., K
j 1 ……(1)
 Note that these equations have the same from as the set of
equations that define the stationary pmf for a discrete-time
Markow chain with transition probabilities Pjk.
 Close Queuing Networks
The only difference is that the sum of the k ‘s need not be
one. Thus the solution vector to eq. (1) must be
proportional to the stationary pmf j corresponding to
{Pjk} :
K
k   ( I ) k , where  k    j Pjk
j 1 ……(2)

And where (I) is a constant that depends on I, the

number of customers in the queuing network. If we sum
both sides of eq. (2) over k, we see that (I) is the sum of
the arrival rates in all the queues un the network, and k
= k / (I) is the fraction of the total arrivals to queue k.
 Close Queuing Networks
Theorem: Let lk = (I) k be a solution to eq.(1),
and let n = (n1, n2, …,nk) by any state of the
network for which n1, n2, …,nk  0 and
n1  n2  ...  nK  I .....(3)
Then
P[ N1  n1 ]P[ N 2  n2 ]...P[ N K  n K ]
P[ N (t )  n]  , .....(4)
S (I )

where P[ Nk = nk] is the steady state pmf of an M/M/ck

system with arrival rate k and service rate k, and where
S(I) is the normalization constant given by
 Close Queuing Networks
S (I )  
n:n1 ... n K 1
P[ N1  n1 ]P[ N 2  n2 ]...P[ N K  n K ], .....( 5)

Equation 4 states that P[ N(t) = n] has a product form.

However P[ N(t) = n] is no longer equal to the product
of the marginal pmf ’s because of the normalization
constant S(I).
This constant arises because the fact that there are
always I customers in the network implies that the
allowable states n must satisfy equation (3).
The theorem can be proved by taking the approach used
to prove Jackson ‘s theorem.
 Close Queuing Networks
Example: Suppose the computer system example that we have
seen in open queuing network is operated so that there are always
I programs in the system. The resulting network of queue is shown
below. Note that the feedback loop around the CPU signifies the
completion of the one job and its instantaneous replacement by
another one. Find the steady state pmf of the system. Find the rate
at which programs are completed.

CPU 
p
 
 Close Queuing Networks
 Solution: The stationary probabilities associated with equation (1) are
found by solving

 1  p 1   2,  2  (1  p) 1, and 1  1  1
 The stationary probabilities are then
1 1 p
1  and 2 
2 p 2 p
And the arrival rates are

 (I ) (1  p ) ( I )
1   ( I ) 1  and 2 
2 p 2 p
Solution Continues…
 Close Queuing Networks
 The stationary pmf for the network is then
(1  1 ) 1i (1   2 )  2I i
P[ N1  i, N 2  I  i ]  0i I
S (I )
I
and S ( I )  (1  1 )(1   2 ) 1i  2I i
i 0

1  i
So P[ N1  i, N 2  I  i]   0i I
1  I 1

1  1 2 2 1 2
where     and 1  ,  2 
 2  2 1 (1  p) 1 1 2

Solution Continues…
 Close Queuing Networks
 The rate at which programs are completed is pl1 . We find l1 from
the relation between server utilization and probability of an
empty system:
1 1 
1  P[ N1  0] 
1 1   I 1
which implies that
 (1   I )
p1  p1
1  I 1